Data readiness is now a supplier-qualification criterion. Can your plant prove what it makes?
A peer note on the quiet change in how orders get allocated — and why the winning plants fixed their data before they touched AI.
Run a mid-market plant in India and your week has a familiar shape. The quote desk grinds through RFQs. QA preps the next customer audit. Export paperwork takes longer than the production did. And the AI noise hums in the background — announcements from companies fifty times your size. A board member asks about the plan.
In most plants we walk into, the honest answer is: watching. It has felt like prudence.
Then.
The qualification packet changed. Global OEMs now ask suppliers for traceability by lot. For near-zero-defect evidence. For 8D reports backed by data, not memory. ACMA’s president said it without decoration in February: smart factories are no longer optional. The China-plus-one order goes to whoever passes the data audit. Not whoever quotes lowest.
ACMA · February 2026
The border changed too. CBAM went definitive on 1 January. A plant with no verified emissions data gets assessed at default values — the highest ones. GTRI puts the damage at 15–22% of price. Not for polluting more. For measuring less. And this week, the EU’s Digital Product Passport registry goes live for textiles. Sector by sector, the pattern repeats.
GTRI · CBAM · Jan 2026 · EU DPP · Jul 2026
The audit moved from your parts to your data.
Now the strange part. India’s large manufacturers discuss AI as an operating fact in their AGMs. The mid-market is close to silent. That silence used to read as discipline — let the big companies burn the early money. It now reads as exposure. The audit doesn’t scale down for company size.
Affordability is the standard excuse. It no longer holds. Instrumenting one seam of the value chain costs less than one lost order. And the government has moved: a 15% capital subsidy, plus a ₹50,000-per-unit MSME pilot for AI hardware and analytics.
The constraint was never capex. It’s a named owner and a place to start.
Three production systems showed us where that place is. None started on the shop floor.
A precision fastener maker.
An Indian supplier to auto OEMs. A backlog of 500 RFQs — each quote needed one of three senior people. We built drawing-to-costed-quote and ran it live. The backlog cleared. Quotes now leave in under 48 hours. The hard part wasn’t the model. It was 12-plus hours of interviews per domain, turning why this machine, why this price into rules the system must pass. Senior judgement, made checkable.
A German precision manufacturer.
German. Brownfield. Machines that don’t want to talk — Euromap and OPC UA where they exist, small gateways where they don’t. The system plans machine selection from five years of routing history. AI plans the familiar 70% of jobs. The veterans keep the exceptions. And nothing went live before weeks of shadow mode against real decisions. Trust was earned on evidence.
A global components group.
Supplier risk agents across a 30,000-supplier base. Certificate intelligence across plants, in multiple languages — the paper layer where audits are won and lost. All of it read-only on SAP. Zero footprint on the systems of record.
Same pattern everywhere: the systems that survive start at the seams of the value chain — the quote desk, the certificate pile, the planning board. Where the data already exists and the money already leaks. Not with a lighthouse project.
The value chain pays before the shop floor.
The lighthouse endpoint is real. Tata Steel disclosed 860-plus AI models live at its AGM this month. Two-thirds of surveyed component makers run smart-factory tech, per ACMA — reporting 10–20% productivity and 20–30% quality gains. But all of it sits downstream of a quieter asset. Data that survives someone else’s scrutiny.
Tata Steel AGM · Jul 2026 · ACMA survey · Feb 2026
Verified data is now a commercial asset. It prices your parts at the border. It passes your customer’s audit. It wins the China-plus-one order.
The playbook from the field runs four steps, in order — whoever runs them.
01
Start at the seams, not the showcase.
Quote desk, certificates, planning, test reports. Data exists. Money leaks. Results count in weeks.
Proven at a fastener maker
02
Capture judgement as checkable rules.
Thirty years of “why this machine” becomes constraints the system must pass. The veterans keep the exceptions.
Proven in a German brownfield plant
03
Shadow mode before go-live.
Weeks of parallel running against real decisions, scored blind. The plant trusts what it has watched work.
Standard across engagements
04
Name the owner, then train through reps.
Process owners lead adoption — Mahindra’s rule, and it matches everything we’ve seen. Timed practice on your own jobs, until the new way is the way.
Proven across sectors
The first ninety days are concrete. Weeks one to three: walk the flow from RFQ to dispatch and score each seam — is the data there, is money leaking? Weeks four to ten: instrument one seam, interviews first, shadow mode from day one. Weeks eleven to thirteen: measure the before-and-after on real work, then pick seam two from evidence. One named owner throughout. Mahindra runs 100% AI weld-testing and reports 10–15% more uptime — with process owners, not IT, leading.
Month six is a different plant.
If you’re holding an RFQ backlog, a customer data mandate, or a CBAM exposure sheet, the question isn’t whether to “do AI”. It’s which seam gets instrumented first — and who owns it by name.
That’s a design decision. Your next audit will grade it either way.
Happy to compare notes — no agenda beyond it. The people I learn most from are running the same problem on their own floor.
